I am Not Dead Yet: Identification of False-Positive Matches to Death Master File

Area of Science:

• Health Informatics
• Clinical Data Management
• Biostatistics

Background:

• Accurate patient death ascertainment is crucial for clinical outcomes.
• Current methods using database matching with external death indices have imperfect accuracy.
• False positive matches to death records can lead to significant data errors.

Purpose of the Study:

• To investigate the utility of clinical records created after the date of death.
• To determine if these post-death records can identify false positive matches to the Death Master File.
• To assess the potential for improving existing record linkage algorithms.

Main Methods:

• Analysis of clinical data recorded more than one month after the date of death.
• Evaluation of positive predictive value (PPV) and sensitivity of different data types (notes, labs, billing, vital signs, medications).
• Development of a multivariable logistic model incorporating post-death recorded data elements.

Main Results:

• PPV ranged from 74.7% (notes) to 95.9% (labs); sensitivity ranged from 57.4% (adverse medication reactions) to 94.9% (notes).
• Combining any two of four data elements (billing, labs, vital signs, medications) yielded 83.0% sensitivity and 98.3% PPV.
• A logistic model using post-death data elements achieved an ROC curve area of 0.987.

Conclusions:

• Clinical data recorded after the date of death effectively identifies false positive matches to death indices.
• This approach can significantly enhance the accuracy of record linkage algorithms.
• Utilizing post-death clinical data offers a promising strategy for improving patient death ascertainment.